→

How to Switch from ChatGPT to Claude

Almost nothing about this migration is technical. Both products are a text box in a browser, you will be productive in the first minute, and the export takes ten. What makes people bounce back to ChatGPT is subtler: two years of conversation history stays behind, the custom GPTs someone built for their team have no direct import path, and the workflows that depended on image generation or a specific integration quietly stop working. None of that is a reason not to switch — it's a reason to spend an hour on the parts that don't move before you cancel anything. This guide covers the export, what to rebuild as Projects, the capabilities that genuinely have no equivalent, and the cases where staying is the right answer.

General-purpose AI assistants⏱ About an hour, plus a week of noticing which habits don't transfer7-step migration

Why people leave ChatGPT

📄 Long documents handled as documents

A large context window means a full contract, a research paper set, or a sprawling codebase can go in whole rather than in chunks you stitch back together. If your work is 'read all of this and tell me what's in it', this is the difference that actually changes your day.

✍️ Writing that needs less rewriting

The most common reason people move is prose quality — output that reads less like a template and holds a register you asked for across a long piece. This is subjective and you should test it on your own writing rather than take anyone's word for it, but it's the reason the switch sticks when it sticks.

🗂️ Projects as persistent working context

A Project holds your reference documents and standing instructions so every conversation inside it starts already knowing your style guide, your schema, or your client's brief. It's the closest analogue to a custom GPT and it's the thing to rebuild first.

🧩 Artifacts you can keep editing

Generated documents, code and small apps open in a side panel you iterate on rather than a wall of chat text you copy out. For anything you're producing rather than just asking about, that's a materially better surface.

Stay on ChatGPT if…

A migration you regret costs more than the subscription you were trying to escape. These are the cases where ChatGPT is still the right answer.

•You use image generation regularly. It's built into ChatGPT and it is not part of Claude. If 'make me an image' is a routine part of your week, switching means adding a separate image tool, and you should price that in before you cancel rather than discover it mid-task.
•Your team runs on custom GPTs that other people use. Those are a shared artefact with an audience — rebuilding them as Projects is straightforward for you and invisible to everyone who had them bookmarked. Migrate the team, not just yourself, or don't migrate.
•You depend on ChatGPT's voice mode, or on a specific integration or connector that your workflow is built around. Check the equivalent exists on the other side first; 'roughly similar feature set' is not the same as 'the thing you actually use every day'.
•Your organisation has approved ChatGPT through security review and Claude is not on the list. That's a procurement conversation, not a tooling preference — start it properly rather than moving company data to an unreviewed vendor.

What you gain and what you give up

Capability	ChatGPT	Claude
Long-document analysis	Good	Larger context, fewer chunking workarounds
Long-form writing quality	Strong	Often preferred for register and restraint
Persistent project context	Custom GPTs and memory	Projects with attached knowledge
Editable generated artefacts	Canvas	Artifacts
Image generation	Built in	Not available
Voice conversation	Mature	Limited
Shareable custom assistants	Custom GPTs with a directory	Projects, shared within a team
Third-party integrations and connectors	Broader ecosystem	Narrower but growing
Enterprise approval status	Widely approved already	Often needs a fresh review

Green marks the stronger side of each row. Rows where both are comparable are left neutral.

The migration, step by step

Export your ChatGPT data while the account is still live

10 minutes, plus waiting for the email

Request the data export from settings. It arrives as an archive by email and contains your conversation history in a readable form. You cannot import it into Claude — nothing does — but you can search it, and the archive is the only copy of two years of context you will ever get. Do this before you cancel anything, because the option disappears with the account.

Copy your custom instructions out by hand

10 minutes

Open your custom instructions and paste them into a plain text file. These are usually a few paragraphs describing how you want to be talked to and what you're working on, they took longer to get right than you remember, and there is no export button for them. They become the standing instructions on your first Project.

Inventory the custom GPTs you actually use

20 minutes

List the custom GPTs you open more than once a month, and for each one copy out its instructions and note which files it had attached. Most people find the real list is two or three, not the dozen they built. Anything with other people using it needs a heads-up before it goes away — that's the part that turns a personal migration into a team problem.

Rebuild your top workflows as Projects

30 minutes

Create a Project per recurring context — one per client, one per codebase, one for your own writing — and put the instructions and reference documents you just collected into each. Starting conversations inside a Project rather than in a blank chat is the habit that makes the new tool feel like it knows you, and it's the step that decides whether the migration sticks.

Run the same real task through both, side by side

30 minutes

Take a piece of work you did last week and run it through both tools with the same prompt. Not a benchmark question — your actual work, with your actual documents. Prose preference is genuinely subjective and this is the only test that answers it for you. Keep both tabs open while you do it.

Find the replacement for anything that didn't come across

20 minutes

If image generation, voice, or a specific connector was part of your routine, decide now what fills the gap and whether it's worth the extra subscription. This is the arithmetic that determines whether switching actually saves money — a cheaper assistant plus a second tool for the thing it doesn't do may not be cheaper at all.

Overlap for one billing cycle, then cancel deliberately

5 minutes at the end of the month

Pay for both for a month. It costs one subscription and it removes the pressure to justify a decision you've already paid for. At the end, cancel the one you stopped opening — and if that's the new one, that's a successful trial, not a failure.

What does not come across

Every migration loses something. Budget for these before you cancel ChatGPT, not after.

✗Your entire conversation history. It exports for your own reference and imports nowhere — searchable history restarts on migration day.
✗Custom GPTs, including any you shared with colleagues or published. Projects are the analogue, but the links people bookmarked will not work.
✗Image generation as part of the same subscription.
✗Voice conversation, if that was part of how you actually used the product.
✗Any third-party integration or connector without an equivalent on the other side — check your specific ones rather than assuming parity.

ChatGPT → Claude: common questions

Can I import my ChatGPT conversation history into Claude?

No. You can export it from ChatGPT for your own records, but there is no import path on the other side and no third-party tool that meaningfully reconstructs it. Treat the archive as a searchable reference and accept that in-product history starts fresh — which is exactly why you export before cancelling, not after.

What replaces my custom GPTs?

Projects. You create one per recurring context, paste in the instructions and attach the same reference files, and start conversations inside it. The rebuild is quick for a personal GPT; the part to plan for is other people — anyone who was using a GPT you built loses access to it, and no export migrates that audience for you.

Does Claude generate images?

No. That's the clearest capability gap in this migration. If image generation is part of your routine you'll need a separate tool for it, and you should decide which one and what it costs before you cancel — this is the single most common reason people quietly go back.

Is Claude actually better at writing than ChatGPT?

Many people find its long-form prose needs less editing and holds a requested register better, which is the most common stated reason for switching. But this is a preference, not a benchmark, and it varies by the kind of writing you do. Run one real piece of your own work through both before you decide — it takes fifteen minutes and it's the only evidence that applies to you.

Should I keep paying for both?

For one month, yes — deliberately. Beyond that it's only worth it if you're genuinely using each for different work, most commonly keeping ChatGPT for image generation and voice while writing and document analysis moves across. If you can't name what each one is for, you're paying twice for one habit.